Schematic Conclusions of Medical Study
This is not the end of the analysis. So far, we have assumed that each of the curves to be totally accurate. In fact, curves in typical clinical trials will have significant statistical errors. If each population (those who took FactStatin, and those who took a placebo) was 5000 people at the start, then there will be 400 people (8%) in the placebo group and 300 people (6%) in the FactStatin group who have CHD. Are these number statistically different?
In fact, they are! However, there is more to this story...
People drop out during such studies. Sometimes they don't like the side effects. Sometimes they get tired of the persistent monitoring. Sometimes they get an unrelated illness. Sometimes they just decide to leave for no obvious reason.
It is not unusual for studies to terminate with less that half their original participants. Therefore the the actual numbers of participants who have CHD are significantly smaller, making the result less precise. The Kaplan-Meier process compensates for these "dropouts" (it's called "censoring") but the negative impact on the precision cannot be mitigated.
The Risk Ratio (RR - also known as Relative Risk or Odds Ratio) - this is the ratio that shows "how much better off" you are using the medication. In our case, it would be 300/400 (=0.75). Your chance of getting CHD at the end date of the study is less, when you take the medication - it has gone down by 25%.
The RR will be accompanied by a range, the Margin of Error, representing the statistical measurement error. This provides an estimate of the range within which, the "true" result is expected to be. (See the statistics tutorial for more information. In brief, it means that if you were to repeat this experiment many times, 95% of the times you would get a result in this range!). A typical result, given our numbers, would be 0.75 ± 0.10
Possibly the most important result that sometimes gets less press, is the absolute risk. It tells one what the actual odds (probability) of getting the illness over a predetermined period are. For example, some heart disease calculators will provide a 10 year risk. Let's say its 6%, for someone using the calculator. What this means is that if we took a group of 100 people, all with the same characteristics, 6 of them would be expected to get heart disease over the next 10 years.